src/gallium/drivers/lima/lima_program.c

   1 /*
   2  * Copyright (c) 2017-2019 Lima Project
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sub license,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the
  12  * next paragraph) shall be included in all copies or substantial portions
  13  * of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  21  * DEALINGS IN THE SOFTWARE.
  22  *
  23  */
  24
  25 #include "util/u_memory.h"
  26 #include "util/ralloc.h"
  27 #include "util/u_debug.h"
  28
  29 #include "tgsi/tgsi_dump.h"
  30 #include "compiler/nir/nir.h"
  31 #include "nir/tgsi_to_nir.h"
  32
  33 #include "pipe/p_state.h"
  34
  35 #include "lima_screen.h"
  36 #include "lima_context.h"
  37 #include "lima_program.h"
  38 #include "lima_bo.h"
  39 #include "ir/lima_ir.h"
  40
  41 static const nir_shader_compiler_options vs_nir_options = {
  42    .lower_ffma = true,
  43    .lower_fpow = true,
  44    .lower_ffract = true,
  45    .lower_fdiv = true,
  46    .lower_fmod = true,
  47    .lower_fsqrt = true,
  48    .lower_sub = true,
  49    .lower_flrp32 = true,
  50    .lower_flrp64 = true,
  51    .lower_ftrunc = true,
  52    /* could be implemented by clamp */
  53    .lower_fsat = true,
  54    .lower_bitops = true,
  55    .lower_rotate = true,
  56    .lower_sincos = true,
  57 };
  58
  59 static const nir_shader_compiler_options fs_nir_options = {
  60    .lower_ffma = true,
  61    .lower_fpow = true,
  62    .lower_fdiv = true,
  63    .lower_fmod = true,
  64    .lower_sub = true,
  65    .lower_flrp32 = true,
  66    .lower_flrp64 = true,
  67    .lower_fsign = true,
  68    .lower_rotate = true,
  69    .lower_fdot = true,
  70    .lower_bitops = true,
  71    .lower_vector_cmp = true,
  72 };
  73
  74 static const struct nir_lower_tex_options tex_options = {
  75    .lower_txp = ~0u,
  76 };
  77
  78 const void *
  79 lima_program_get_compiler_options(enum pipe_shader_type shader)
  80 {
  81    switch (shader) {
  82    case PIPE_SHADER_VERTEX:
  83       return &vs_nir_options;
  84    case PIPE_SHADER_FRAGMENT:
  85       return &fs_nir_options;
  86    default:
  87       return NULL;
  88    }
  89 }
  90
  91 static int
  92 type_size(const struct glsl_type *type, bool bindless)
  93 {
  94    return glsl_count_attribute_slots(type, false);
  95 }
  96
  97 void
  98 lima_program_optimize_vs_nir(struct nir_shader *s)
  99 {
 100    bool progress;
 101
 102    NIR_PASS_V(s, nir_lower_viewport_transform);
 103    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
 104    NIR_PASS_V(s, nir_lower_load_const_to_scalar);
 105    NIR_PASS_V(s, lima_nir_lower_uniform_to_scalar);
 106    NIR_PASS_V(s, nir_lower_io_to_scalar,
 107               nir_var_shader_in|nir_var_shader_out);
 108
 109    do {
 110       progress = false;
 111
 112       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 113       NIR_PASS(progress, s, nir_lower_alu_to_scalar, NULL);
 114       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 115       NIR_PASS(progress, s, nir_copy_prop);
 116       NIR_PASS(progress, s, nir_opt_remove_phis);
 117       NIR_PASS(progress, s, nir_opt_dce);
 118       NIR_PASS(progress, s, nir_opt_dead_cf);
 119       NIR_PASS(progress, s, nir_opt_cse);
 120       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 121       NIR_PASS(progress, s, nir_opt_algebraic);
 122       NIR_PASS(progress, s, nir_opt_constant_folding);
 123       NIR_PASS(progress, s, nir_opt_undef);
 124       NIR_PASS(progress, s, nir_opt_loop_unroll,
 125                nir_var_shader_in |
 126                nir_var_shader_out |
 127                nir_var_function_temp);
 128    } while (progress);
 129
 130    NIR_PASS_V(s, nir_lower_int_to_float);
 131    NIR_PASS_V(s, nir_lower_bool_to_float);
 132
 133    /* Some ops must be lowered after being converted from int ops,
 134     * so re-run nir_opt_algebraic after int lowering. */
 135    do {
 136       progress = false;
 137       NIR_PASS(progress, s, nir_opt_algebraic);
 138    } while (progress);
 139
 140    NIR_PASS_V(s, nir_copy_prop);
 141    NIR_PASS_V(s, nir_opt_dce);
 142    NIR_PASS_V(s, nir_lower_locals_to_regs);
 143    NIR_PASS_V(s, nir_convert_from_ssa, true);
 144    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 145    nir_sweep(s);
 146 }
 147
 148 void
 149 lima_program_optimize_fs_nir(struct nir_shader *s)
 150 {
 151    BITSET_DECLARE(alu_lower, nir_num_opcodes) = {0};
 152    bool progress;
 153
 154    BITSET_SET(alu_lower, nir_op_frcp);
 155    BITSET_SET(alu_lower, nir_op_frsq);
 156    BITSET_SET(alu_lower, nir_op_flog2);
 157    BITSET_SET(alu_lower, nir_op_fexp2);
 158    BITSET_SET(alu_lower, nir_op_fsqrt);
 159    BITSET_SET(alu_lower, nir_op_fsin);
 160    BITSET_SET(alu_lower, nir_op_fcos);
 161
 162    NIR_PASS_V(s, nir_lower_fragcoord_wtrans);
 163    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
 164    NIR_PASS_V(s, nir_lower_regs_to_ssa);
 165    NIR_PASS_V(s, nir_lower_tex, &tex_options);
 166
 167    do {
 168       progress = false;
 169
 170       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 171       NIR_PASS(progress, s, nir_lower_alu_to_scalar, alu_lower);
 172       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 173       NIR_PASS(progress, s, nir_copy_prop);
 174       NIR_PASS(progress, s, nir_opt_remove_phis);
 175       NIR_PASS(progress, s, nir_opt_dce);
 176       NIR_PASS(progress, s, nir_opt_dead_cf);
 177       NIR_PASS(progress, s, nir_opt_cse);
 178       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 179       NIR_PASS(progress, s, nir_opt_algebraic);
 180       NIR_PASS(progress, s, nir_opt_constant_folding);
 181       NIR_PASS(progress, s, nir_opt_undef);
 182       NIR_PASS(progress, s, nir_opt_loop_unroll,
 183                nir_var_shader_in |
 184                nir_var_shader_out |
 185                nir_var_function_temp);
 186    } while (progress);
 187
 188    NIR_PASS_V(s, nir_lower_int_to_float);
 189    NIR_PASS_V(s, nir_lower_bool_to_float);
 190
 191    /* Some ops must be lowered after being converted from int ops,
 192     * so re-run nir_opt_algebraic after int lowering. */
 193    do {
 194       progress = false;
 195       NIR_PASS(progress, s, nir_opt_algebraic);
 196    } while (progress);
 197
 198    /* Must be run after optimization loop */
 199    NIR_PASS_V(s, lima_nir_scale_trig);
 200
 201    /* Lower modifiers */
 202    NIR_PASS_V(s, nir_lower_to_source_mods, nir_lower_all_source_mods);
 203    NIR_PASS_V(s, nir_copy_prop);
 204    NIR_PASS_V(s, nir_opt_dce);
 205
 206    NIR_PASS_V(s, nir_lower_locals_to_regs);
 207    NIR_PASS_V(s, nir_convert_from_ssa, true);
 208    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 209
 210    NIR_PASS_V(s, nir_move_vec_src_uses_to_dest);
 211    NIR_PASS_V(s, nir_lower_vec_to_movs);
 212
 213    nir_sweep(s);
 214 }
 215
 216 static void *
 217 lima_create_fs_state(struct pipe_context *pctx,
 218                      const struct pipe_shader_state *cso)
 219 {
 220    struct lima_context *ctx = lima_context(pctx);
 221    struct lima_screen *screen = lima_screen(pctx->screen);
 222    struct lima_fs_shader_state *so = rzalloc(NULL, struct lima_fs_shader_state);
 223
 224    if (!so)
 225       return NULL;
 226
 227    nir_shader *nir;
 228    if (cso->type == PIPE_SHADER_IR_NIR)
 229       nir = cso->ir.nir;
 230    else {
 231       assert(cso->type == PIPE_SHADER_IR_TGSI);
 232
 233       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 234    }
 235
 236    lima_program_optimize_fs_nir(nir);
 237
 238    if (lima_debug & LIMA_DEBUG_PP)
 239       nir_print_shader(nir, stdout);
 240
 241    if (!ppir_compile_nir(so, nir, screen->pp_ra, &ctx->debug)) {
 242       ralloc_free(so);
 243       return NULL;
 244    }
 245
 246    return so;
 247 }
 248
 249 static void
 250 lima_bind_fs_state(struct pipe_context *pctx, void *hwcso)
 251 {
 252    struct lima_context *ctx = lima_context(pctx);
 253
 254    ctx->fs = hwcso;
 255    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_FRAG;
 256 }
 257
 258 static void
 259 lima_delete_fs_state(struct pipe_context *pctx, void *hwcso)
 260 {
 261    struct lima_fs_shader_state *so = hwcso;
 262
 263    if (so->bo)
 264       lima_bo_free(so->bo);
 265
 266    ralloc_free(so);
 267 }
 268
 269 bool
 270 lima_update_vs_state(struct lima_context *ctx)
 271 {
 272    struct lima_vs_shader_state *vs = ctx->vs;
 273    if (!vs->bo) {
 274       struct lima_screen *screen = lima_screen(ctx->base.screen);
 275       vs->bo = lima_bo_create(screen, vs->shader_size, 0);
 276       if (!vs->bo) {
 277          fprintf(stderr, "lima: create vs shader bo fail\n");
 278          return false;
 279       }
 280
 281       memcpy(lima_bo_map(vs->bo), vs->shader, vs->shader_size);
 282       ralloc_free(vs->shader);
 283       vs->shader = NULL;
 284    }
 285
 286    return true;
 287 }
 288
 289 bool
 290 lima_update_fs_state(struct lima_context *ctx)
 291 {
 292    struct lima_fs_shader_state *fs = ctx->fs;
 293    if (!fs->bo) {
 294       struct lima_screen *screen = lima_screen(ctx->base.screen);
 295       fs->bo = lima_bo_create(screen, fs->shader_size, 0);
 296       if (!fs->bo) {
 297          fprintf(stderr, "lima: create fs shader bo fail\n");
 298          return false;
 299       }
 300
 301       memcpy(lima_bo_map(fs->bo), fs->shader, fs->shader_size);
 302       ralloc_free(fs->shader);
 303       fs->shader = NULL;
 304    }
 305
 306    ctx->pp_max_stack_size = MAX2(ctx->pp_max_stack_size, ctx->fs->stack_size);
 307
 308    return true;
 309 }
 310
 311 static void *
 312 lima_create_vs_state(struct pipe_context *pctx,
 313                      const struct pipe_shader_state *cso)
 314 {
 315    struct lima_context *ctx = lima_context(pctx);
 316    struct lima_vs_shader_state *so = rzalloc(NULL, struct lima_vs_shader_state);
 317
 318    if (!so)
 319       return NULL;
 320
 321    nir_shader *nir;
 322    if (cso->type == PIPE_SHADER_IR_NIR)
 323       nir = cso->ir.nir;
 324    else {
 325       assert(cso->type == PIPE_SHADER_IR_TGSI);
 326
 327       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 328    }
 329
 330    lima_program_optimize_vs_nir(nir);
 331
 332    if (lima_debug & LIMA_DEBUG_GP)
 333       nir_print_shader(nir, stdout);
 334
 335    if (!gpir_compile_nir(so, nir, &ctx->debug)) {
 336       ralloc_free(so);
 337       return NULL;
 338    }
 339
 340    return so;
 341 }
 342
 343 static void
 344 lima_bind_vs_state(struct pipe_context *pctx, void *hwcso)
 345 {
 346    struct lima_context *ctx = lima_context(pctx);
 347
 348    ctx->vs = hwcso;
 349    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_VERT;
 350 }
 351
 352 static void
 353 lima_delete_vs_state(struct pipe_context *pctx, void *hwcso)
 354 {
 355    struct lima_vs_shader_state *so = hwcso;
 356
 357    if (so->bo)
 358       lima_bo_free(so->bo);
 359
 360    ralloc_free(so);
 361 }
 362
 363 void
 364 lima_program_init(struct lima_context *ctx)
 365 {
 366    ctx->base.create_fs_state = lima_create_fs_state;
 367    ctx->base.bind_fs_state = lima_bind_fs_state;
 368    ctx->base.delete_fs_state = lima_delete_fs_state;
 369
 370    ctx->base.create_vs_state = lima_create_vs_state;
 371    ctx->base.bind_vs_state = lima_bind_vs_state;
 372    ctx->base.delete_vs_state = lima_delete_vs_state;
 373 }